The Firm Had an AI Policy: Johnson v. Dunn and the Gap Where Rule 11 Does Not Reach

What the court actually ordered

Seven items, and only one resembles a conventional sanction. Public reprimand, publication in the Federal Supplement, disqualification from the case, a list of every jurisdiction where the attorneys are licensed, referral to the Alabama State Bar and other licensing authorities, and service of the order on clients, opposing counsel and judges in every pending case. No monetary fine ^[1].

The remedies are worth listing in full, because the shape of them is the point. The court publicly reprimanded three attorneys, Matthew B. Reeves, William J. Cranford, and William R. Lunsford, for the misconduct described in the order ^[1]. The sanctions order is also available in accessible form through contemporaneous coverage for readers who would rather not work from the reporter PDF ^[4].

To effectuate that reprimand, it ordered them to provide a copy of the order to their clients, opposing counsel, and the presiding judge in every pending state or federal case in which they were counsel of record, and to every attorney in their law firm, within ten days, certifying compliance within twenty-four hours of completing it ^[1].

It directed the Clerk to submit the order for publication in the Federal Supplement, disqualified all three from further participation in the case, required them to provide a listing of every jurisdiction in which they are licensed within twenty-four hours, and directed the Clerk to serve the order on the General Counsel of the Alabama State Bar and any other applicable licensing authorities ^[1].

The seventh item matters as much as the other six: two other attorneys and the firm itself were released without sanction ^[1]. This was not a decision to punish an organisation. The court located responsibility with specific individuals and said so.

Notice what is absent. There is no fine. Our analysis of what happens when you file a brief with a fake AI citation covers the monetary end of the range, and the figures there are mostly modest. Johnson v. Dunn suggests the monetary figure was never the real exposure.

Why Rule 11 did not apply

Rule 11(d) provides that Rule 11 "does not apply to ... motions under Rules 26 through 37." The filings here were a motion for leave to depose and a motion to compel, both discovery motions under Rules 30 and 37. The court held Rule 11 therefore did not reach them, whatever their content ^[1][2].

This is the part most summaries of the case omit, and it is the part with the widest application. Rule 11 is the certification most lawyers have in mind when they think about their exposure for what a filing says. It is not a general warranty of every paper filed in a case.

The exception is textual and specific. Rule 11(d) states that the rule does not apply to disclosures and discovery requests, responses, objections, and motions under Rules 26 through 37 ^[2]. The 1993 advisory committee note explains the reasoning: Rules 26(g) and 37 establish their own certification standards and sanctions, and it is appropriate that those rules, "specially designed for the discovery process," govern such documents rather than the more general provisions of Rule 11 ^[1].

The court then identified the problem with that division of labour. The committee expected Rules 26 and 37 to allow courts to address sanctionable misconduct in discovery motions, but those rules "furnish no basis for the court to address the false statements of law these attorneys made in discovery motions" ^[1][3]. Rules 26 and 37 are built around discovery conduct, not around fabricated legal authority.

The court named the result plainly. "This appears to be an unintended anomaly in Rule 11," it wrote, and then: "Regardless, Rule 11 says what it says, it does not apply here" ^[1].

The practical consequence is a gap in the mental model most litigators carry. A fabricated citation in a summary judgment brief runs into Rule 11. The same fabricated citation in a motion to compel does not. Nothing about the conduct changed; only the procedural posture did. Our guide to responding when opposing counsel files fabricated citations works through the Rule 11 route, and this case is the reminder that the route is not always open.

Rule 11(d) is a federal rule. State procedural rules differ, and some have no equivalent exception. This section describes what the Johnson v. Dunn court held about the federal rule in the case before it; confirm the position in your own jurisdiction.

Inherent authority: what fills the gap, and what constrains it

With Rule 11 unavailable, the court proceeded under its inherent power. Quoting Chambers v. NASCO, it noted that while Rule 11 "reaches only certain individuals or conduct, the inherent power extends to a full range of litigation abuses," but that inherent powers "must be exercised with restraint and discretion" and require a finding of bad faith ^[1].

Inherent authority is the residual power a court holds to manage proceedings before it. It is broader than any particular rule and, precisely because it is broader, hedged with more caution.

The opinion sets out the sequence a court is meant to follow. Where bad-faith conduct could be adequately sanctioned under the Rules, a court ordinarily should rely on the Rules rather than the inherent power. But "if in the informed discretion of the court, neither the statute nor the Rules are up to the task, the court may safely rely on its inherent power" ^[1]. Having found Rule 11 inapplicable and Rules 26 and 37 unable to reach false statements of law, the court concluded the Rules were not up to the task.

The trade-off is the standard of proof. Rule 11 imposes an objective standard of reasonable inquiry and does not require a finding of bad faith. Inherent-power sanctions do. That is a meaningfully higher bar, and it is the reason the court's findings dwell at length on each attorney's state of mind rather than simply on the fact of the fabrication.

For a firm, the asymmetry cuts in an unexpected direction. The route with the lower threshold, Rule 11, carries the safe harbour and a deterrence-capped remedy. The route with the higher threshold, inherent authority, has no safe harbour and a remedy limited only by what the court considers necessary to vindicate its authority. A lawyer who reasons that discovery motions carry less exposure because Rule 11 does not apply has the analysis backwards.

The AI policy did not save them, and the court said why

The court's aggravating findings are directed at exactly the firm profile this site usually addresses. The attorneys "benefitted from repeated warnings, internal controls, and firm policies about the dangers of AI misuse" and had "regular access to gold-standard legal research databases." That context made the lapse worse, not better ^[1].

It would be comfortable to read AI-sanctions cases as stories about under-resourced practitioners without governance. This one is the opposite, and it is worth sitting with rather than filing away.

The court described three well-trained, experienced attorneys at "a large, high-functioning, well-regarded law firm," who "benefitted from repeated warnings, internal controls, and firm policies about the dangers of AI misuse," who had "regular access to gold-standard legal research databases," and who "must have known they would be deeply embarrassed in this kind of situation." Its summary was four words: "And yet here we are" ^[1].

Then the sentence that should be read by anyone who has just finished writing an AI policy: "The reality that this lapse in judgment presented in the most spectacularly unforced fashion underscores the need for more than a fine and reprimand" ^[1]. The existence of the policy did not mitigate. It raised what the court expected, and the gap between the expectation and the conduct is what drove the remedy past a fine.

This complicates advice this site has given, including in our own guide to building a defensible law firm AI policy. A policy remains worth having, and firms without one are worse off. But a policy is not a liability shield, and on this court's reasoning it can function as the opposite: evidence of what the lawyer had already been told.

The distinction that survives is between a policy that informs and a control that binds. Repeated warnings are information; the lawyers here had them. What was missing was anything in the workflow that made an unverified citation hard to file. That is the difference between telling people to verify and building a step they cannot skip, which is the argument for making verification a recorded byproduct of the work rather than an instruction people carry in their heads.

"Recklessness tantamount to bad faith"

Because inherent-power sanctions require bad faith, the court had to locate it. It found that an attorney who drafted, signed and filed motions containing another attorney's fabricated citations acted with "recklessness tantamount to bad faith" in failing to ensure the statements were true, and that a second attorney whose name appeared in the signature block acted with "utter disregard" for their truth ^[1].

The standard matters because it explains how a court reaches bad faith without finding that anyone intended to deceive. Nobody in this case is described as having set out to mislead the court.

The findings distinguish between roles in a way that should worry supervising lawyers. One attorney inserted the false citations. Another drafted, signed and personally filed motions containing them. A third had his name in the signature block. Each was found to have acted with recklessness tantamount to bad faith, on distinct facts ^[1].

The signature-block finding is the one with the widest reach. An attorney whose name appears on a filing he did not draft, containing citations he did not insert, was found to have acted with "utter disregard for [the] truth" of the motions ^[1]. Being the second name on a filing is not a passive status.

That is the same principle the Ninth Circuit reached from a different direction in its point-of-signing decision. Two courts, two procedural vehicles, one conclusion: the signature is where responsibility attaches, and it does not distribute across the people who touched the document.

The consequence class nobody plans for

The requirement to serve the order on clients, opposing counsel and the presiding judge in every pending case, plus every attorney in the firm, is a remedy with no fixed size. Its magnitude depends on how busy the lawyer is. A monetary sanction is bounded at the moment it is imposed; this is not ^[1].

Firms model sanctions exposure as a number. This remedy has no number. Its weight scales with the sanctioned lawyer's active caseload, and for a partner at a large firm that is a great many separate disclosures to clients who had no involvement in the matter.

Consider the mechanics. Every current client receives a court order describing their lawyer's misconduct. Every opposing counsel in every active matter receives the same document, along with whatever tactical use they choose to make of it. Every presiding judge receives it, which means it enters the working knowledge of the bench in front of which that lawyer appears. And every attorney at the firm receives it internally ^[1].

The compliance window compresses it further: ten days to complete, with certification to the court within twenty-four hours of completing ^[1]. There is no version of that timeline that allows the disclosure to be managed quietly.

Publication in the Federal Supplement makes it permanent ^[1]. A monetary sanction is paid and concluded. A published opinion naming three attorneys, quoting their fabricated citations, and describing their firm's policies as having been in place and ignored is a document that will be found by anyone who looks, indefinitely.

The honest planning consequence is that the reputational and relationship exposure is the exposure. Any risk model that treats AI-citation failures as a bounded financial risk is modelling the wrong variable.

Disqualification, and why the court found no hardship

Disqualification usually fails because it punishes the client rather than the lawyer. The court addressed that directly, acknowledging that disqualification "often work[s] substantial hardship on the client," but finding no hardship here because the Attorney General's office had a ready team of capable attorneys, some already on the matter ^[1].

The reason disqualification is rare as a sanction is that its cost falls mainly on someone who did nothing wrong. Courts are reluctant to deprive a party of chosen counsel because of counsel's misconduct.

The court ran that analysis and found it did not bite. At the show cause hearing, the representative of the Alabama Attorney General did not suggest any hardship for the defendant, and the court noted that the Attorney General's office had a ready team of capable attorneys who could step in, some of whom already represented other parties ^[1].

That reasoning contains a conditional worth noticing. Disqualification became available because the client could absorb it. A solo practitioner's client, or a client whose matter depends on one lawyer's accumulated knowledge, presents a different calculation, and the same misconduct might produce a different remedy for reasons that have nothing to do with the lawyer's culpability.

It also means the remedy is least available exactly where the client is most dependent, which is not a comforting symmetry but is the operative one.

What actually mitigated

The court considered and rejected suspension from practice in the district. What persuaded it were things done after the discovery: the firm's internal review, an independent investigation by outside counsel, sincere remorse, and one attorney's commitment to educating others. Response, not policy, is what moved the outcome ^[1].

Having found the pre-existing governance aggravating, the court found the post-discovery response mitigating, and the contrast is the most transferable lesson in the opinion.

It identified two investigative facts specifically: the firm's internal review, and an independent investigation by outside counsel, which together "reassure the court that suspension from the practice of law in the Northern District is not necessary to protect other courts or cases" ^[1]. The firm reviewed its Alabama federal filings and Eleventh Circuit appeals and engaged independent counsel to audit for other erroneous citations.

The court also noted that the lawyers were "sincerely apologetic and remorseful," and that one was "committed to educating others about these matters as a preventive measure" ^[1]. Those are not procedural moves; they are conduct the court evidently found credible.

Read alongside the aggravating findings, the pattern is clear enough to plan around. What a firm had in place before the incident did not help. What it did after the incident kept three lawyers practising in the district. That should shape how a firm designs its response plan, and it is a different emphasis from what most AI-governance advice puts first. Our guide on the AI incident response plan covers the mechanics; this case is the argument for having one.

The independent audit is the specific element worth copying. A firm investigating itself and reporting that it found nothing is a weaker signal than an outside firm doing the same work, and the court's language suggests it treated the two as distinct sources of reassurance.

What to take from this

Four things: Rule 11 does not cover discovery motions and inherent authority is harsher where it applies; a policy raises the standard you are held to rather than shielding you; the signature block is not a passive role; and the response after discovery carries more weight than the governance before it.

None of this depends on Johnson v. Dunn being binding, and it is not binding anywhere. It depends on the reasoning being available to any court facing similar facts, which it plainly is.

The first point is scope. A firm that has trained its lawyers to think about Rule 11 has trained them for one category of filing. Discovery motions sit outside it, and the authority that reaches them is less predictable and less bounded ^[1][2].

The second is about what policies do. They inform, and informed conduct that falls short is judged against what the lawyer was told. If the firm's AI governance consists of warnings and a written policy, it has built something that raises expectations without constraining behaviour.

The third is that verification has to be structural rather than exhortative. The distance between "we told everyone to verify citations" and "an unverified citation cannot reach a filing without a recorded check" is the distance between this case and a firm that never generates it. That is the argument for verification that produces its own record as a byproduct, which our guide on proving verification to a court works through.

The fourth is that the remedy set is wider than money. Disqualification, cascading disclosure to every client and judge, referral to every licensing authority, and permanent publication are all available, and none of them appears in a risk model built around fine amounts.